TSMC and ASML Say AI Chip Demand Will Outrun Supply Through 2028

The two companies at the base of the AI chip supply chain both said this week that demand is still running ahead of what they can build. On October 15, TSMC reported a 69% jump in third-quarter net income and forecast another rise in revenue, with Chairman and CEO C.C. Wei saying AI demand still exceeds the company's supply. A day earlier, ASML said its planned output for 2027 and 2028 will not fully cover customer orders. For enterprise leaders, the message is that compute for AI is likely to stay scarce and expensive for at least the next two years, and planning should account for it.

What did TSMC report?

TSMC's third-quarter earnings release, filed with the SEC on October 15, showed revenue of NT$1.49 trillion, or $46.72 billion, and net income of NT$763.06 billion, up 68.7% from a year earlier, according to CRN Asia. Gross margin came in at 66.4%. Chips made on 7-nanometer and more advanced processes accounted for 79% of wafer revenue, and the newest 2nm node already contributed 8%.

The company guided fourth-quarter revenue to between $52.3 billion and $54.3 billion. According to CRN Asia, that midpoint implies growth of about 58% from a year earlier. High-performance computing, the segment driven by AI accelerators, made up 68% of sales in the quarter. TSMC kept its 2026 capital spending budget at $60 billion to $64 billion and said spending over 2026 to 2028 will be significantly higher than in the previous three years.

Wei told analysts that TSMC's own assessment shows demand far above current supply, and that the company is working to close the gap. He said "all AI players are working with TSMC," CRN Asia reported. He also pointed to a newer source of demand: CPUs and networking chips built for agentic AI, which add to orders for GPUs and other accelerators.

Why ASML's outlook matters as much as TSMC's

ASML makes the lithography machines that TSMC and other chipmakers need to add advanced capacity, so its order book is an early signal of how much new supply is coming. In its Q3 results on October 14, ASML reported net sales of €11.7 billion and net income of €5.0 billion, and forecast full-year 2026 sales of €44 billion to €45 billion.

The more important line was about the future. ASML said planned supply for 2027 and 2028 will not fully meet customer demand. It plans to raise capacity for its main EUV and DUV immersion systems by 30% in 2027 and is studying a further 30% increase for 2028. Some customers, it said, want to place orders for 2029 and 2030. ASML tied that demand mainly to customers' views on how large and how long-lasting AI investment will be.

Investors were not uniformly enthusiastic. Revenue came in slightly below analyst estimates even as profit beat them, and the shares fell about 4% in early trading, XTB reported. TSMC also struck a note of caution, saying consumer and price-sensitive markets face pressure from rising component prices, so it is planning 2027 prudently.

What does a chip shortage mean for enterprise AI budgets?

Most enterprises do not buy wafers or lithography tools. But the constraint passes down the chain to them. When foundry capacity is sold out, hyperscalers and model providers compete for the same accelerators, and that cost shows up in cloud GPU pricing, reserved-capacity contracts and the price of tokens.

That makes efficiency a strategic issue, not just a technical one. At The AI Enterprise Conference 2026, Kaycee Lai, VP of AI at Everpure, argued that the organizations that win with AI will not be the biggest spenders on hardware. "They're not the ones who buy the most GPUs," he said. In his view, the winners are the ones that get their data ready and put it to work. His session on why enterprises are bringing AI to their data also warned that GPUs left idle because storage cannot keep up are a costly problem, since GPUs are the most expensive part of the infrastructure stack.

Other AI Connect speakers have made the cost case from the finance side. At the same conference, Revefi's Girish Bhat and Verisk Global Chief Data Officer Louis Dimodugno described a self-funding AI strategy that uses savings on data platforms to pay for new AI work. Dimodugno warned that token prices could rise once frontier model providers go public. In a supply-limited market, that warning carries more weight.

What to watch next

Two signals will show whether the gap is closing. The first is TSMC's 2nm ramp; the company expects annual 2nm capacity to more than double in 2027, according to CRN Asia. The second is ASML's decision on the extra 2028 capacity.

For CIOs and chief AI officers, the practical step is to treat AI compute as a constrained resource now: lock in capacity for workloads that have proven value, push routine tasks to smaller and cheaper models, and track cost per outcome rather than raw usage. The suppliers are telling the market that relief is not coming soon.

Frequently asked questions

Why does TSMC's earnings report matter to enterprise AI buyers?

TSMC makes most of the advanced chips behind AI services. When its capacity is sold out, the shortage flows through to cloud GPU prices, reserved-capacity contracts and token costs.

What did ASML say about future chip supply?

ASML said its planned supply for 2027 and 2028 will not fully meet customer demand, and some customers want to order for 2029 and 2030.

How should CIOs respond to AI chip supply limits?

Treat compute as a constrained resource: secure capacity for workloads with proven value, use smaller models for routine tasks, and measure cost per business outcome.

Filed under: Data & AI Infrastructure, AI Strategy & ROI